I am interested in quantifying the dispersion of pixel values in one image. More specifically, I want to get an average value of dispersion of the image column-wise (i.e., compute the dispersion of each column of the image and then "average" these values).

Currently, I am using two different metrics: SD and IQR. Regarding the former, the proper way to do so is to average the variances of each column and then take the square root. Is there such particular consideration that should be taken into account for IQR or is it valid to just take the average of the IQR values?

  • $\begingroup$ Never seen a mean IQR before, but in principle it sounds just another particular application of the mean. Wherever I want to use the IQR I am usually troubled by long or fat tails in the raw data and the IQR gives me some resistance already. As a matter of style rather than substance I think I'd expect to see median IQR as well as, if not in preference to, mean IQR. Naturally, if they agree, your position is strengthened. $\endgroup$
    – Nick Cox
    May 19, 2017 at 12:37
  • $\begingroup$ Originally I wanted to use the mean in order to keep it consistent with the SD case. However, what you say indeed makes sense. If there is no reason not to use the mean, I will just report both (mean and median) $\endgroup$ May 24, 2017 at 11:59
  • $\begingroup$ What is this mean intended to represent? What does it tell you? $\endgroup$
    – Glen_b
    Jun 8, 2017 at 5:50


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.